Summary Re: C/Fortran compilers for i860?

A while ago I posted some questions concerning optimizing compilers for
the i860. Here's a summary of the responses I've got so far :


#######################################################################
Reply-To: Larry Meadows <lfm@pgroup.com>


We have been selling compilers for the i860 for 3 years now, both C and
Fortran. Relevant numbers:


Matrix multiply: 27 Mflops
Linpack (DP) 100x100: 9.4 Mflops
Linpack (DP) 300x300: 11.5 Mflops


Spec '89: 33.4


Sorry, don't have SpecFP 92 numbers right now, one customer reported
60+ on an i860XP.


Our customers include:


Intel SSD Meiko (UK) IBM (PVS system, several other cards) Mercury


We have a distributor in Germany, and an office in the UK for European
Tech Support.


For further information, you can fax us at


[US] (503) 682-2806


or call us at


          (503) 682-2637


or send E-mail.


thanks.


Larry Meadows
The Portland Group


lfm@pgroup.com


#######################################################################


Reply-To: cubnt4!pkoehnko@cubnt4.cubenet.sub.org


We have tried the Portland C Compiler and Intel's C compiler (which is,
as far as I know, a OEM version of the Metaware Compiler) and are really
satisfied with the Portland. We got sometimes speed improvments which
are about 2 times faster then the code from GNU-C.


> We're also looking for Fortran compilers. We know of Pacific
> Sierra's VAST, which transforms Fortran programs at the source
> level and inserts calls into a vector library. Are there any
> other optimizing Fortran compilers for the i860?
We also prefer the Portland version of the Fortran compiler. There are
some bugs in the Intelcompiler, related to complex math.


Portland ships for both versions of their compiler a vectorizing version
which are a bit more expensive.


regards
Peter


#######################################################################


Reply-To: bobd@bobd.intel.com (Bob Dietrich)


The Portland Group has both C and Fortran compilers that do an amazing
job optimizing for the i860. Contact:


Denny Cole
Director of Marketing and Sales
9150 SW Pioneer Ct. Suite H
Wilsonville, Oregon 97070
USA
Phone: 503.682.2806
FAX: 503.682.2637
email: ogicse!pgroup!dcole


Bob Dietrich
Intel Corporation, Hillsboro, Oregon
(503) 696-4270 (FAX: 696-4904)
Internet: bobd@bobd.intel.com
or uunet!intelhf!bobd!bobd


#######################################################################


Reply-To: rob@pact.nl (Rob Kurver)


I believe the Portland Group has the best i860 compilers. Try emailing
them at info@pgroup.com.


Cheers. - Rob
--
          PACT Rob Kurver
        Foulkeslaan 87 rob@pact.nl
      2625 RB Delft ph: +31 15 616864
    The Netherlands fax: +31 15 610032


#######################################################################


Reply-To: Paul Puglia <puglia@cucbs.chem.columbia.edu>


Good luck, I was working with a group of researchers who were building
their own i860 processor cards. They developed the software for it using
an alacron (?) AT bus coprocessor card. We tried several of the
commerical compilers available for the i860. We used Metaware, Portland
Groups, and Greenleaf. None of them did too great of a job of instruction
scheduling. This group was developing their hardware to do high precision
numerical calculations and found that on the stuff that was spit out of
the best of the compilers they were only getting about 5 Mflops. They
finally resorted to stopping the compilaton processes at the assembler
stage and hand tuning the resulting assembler. One of the things we were
disappointed in was the inability of most of these compilers to take
advantage of pipeline.


I feel that the lack of compilers is one of the reason that the i860 has
fallen behind the risc offerings from other manufactures.


Paul Puglia
puglia@cucevx.civil.columbia.edu


#######################################################################


Reply-To: miker@metaware.com (Mike Ross)


We would of course be interested in what improvements you might like to
see to the High C compiler in terms of optimization. You didn't say what
version you have. Are you using -Hon=softpipe to get software pipelining?
I should think that would cause your application to use the independent
functional units a lot, unless it just isn't ameniable to parallel
execution.


You may be interested to know that we are working on a FORTRAN compiler,
and will be doing beta testing with it soon. Would you be interested in
being a beta site? It uses the same code generator and optimizer as High
C. If you have an immediate need for a production compiler, you could use
Lahey's FORTRAN, which also uses our backend.


Intel has designated MetaWare as its official compiler for the 860, and
uses it for their internal benchmarking on the chip. I really don't think
there is a higher performance offering on the 860 platform. The other
major vendor is the Portland Group. They do already have a FORTRAN
product, but we've generally beaten them on performance.


If your code isn't running well on the 860, we'd like to see it if
possible, and try to improve what we're doing. Thanks for the opportunity.
Mike Ross




#######################################################################


Reply-To: Kirk Hays <hays@SSD.intel.com>


There's only one set of production quality C and FORTRAN compilers for the
i860, IMHO, and that's the one from the Portland Group, Inc.


Call them in the USA at (503) 682-2806.


I ported/evaluated/productized/criticized their compilers for the iPSC/860
and PARAGON supercomputers for three years, and am just a happy customer,
with no financial interest.


Their customer support is very good.


We see 37 MFLOPS on compiled FORTRAN SAXPY from PGI's compilers, with a
theoretical peak of 40 MFLOPS. This on an i860XR at 40Mhz with a
relatively slow memory subsystem.


#######################################################################


Reply-To: jim@meiko.co.uk (James Cownie)


Matthias,


We use the Portland Group compilers and are generally pretty happy with
them. They produce pretty impressive code from inner loops, with a lot of
instruction scheduling.


Various people at GMD Schloss Birlinghoven have some experience with these
compilers on our machine. You could talk to Clemens Thole for instance.


The sales contact is John Barr (john@pgroup.com) who despite his mail
address is actually in the UK.


-- Jim


James Cownie
Meiko Limited
650 Aztec West
Bristol BS12 4SD
England


Phone : +44 454 616171
FAX : +44 454 618188
E-Mail: jim@meiko.co.uk or jim@meiko.com


#######################################################################


Reply-To: pardo@cs.washington.edu


You might ask Henry Baker at Nimple Computer Corporation, 16231 Meadow
Ridge Way, Encino, CA 91436, (US country code) 818-501-4956, or FAX
818-986-1360. Unfortunately, he is not on the net, or at least wasn't
when last I talked with him about a year ago.


He has been doing work in high-performance compilation for the i860.
Therefore, he is probably familiar with at least some of the tools that
are available. Further, he is himself a compiler writer, has been doing
work in the area of instruction scheduling, etc., so he's likely to be
aware of how good various tools do at those tasks.


Good luck!


;-D on ( Compile Ore ) Pardo


#######################################################################


Reply-To: rstewart@megatek.com (Rich Stewart)


The Portland Group makes a compilers, but if you do alot of integer work,
or all of your code does not vectorize well, you'll be making some
tradeoffs. The problem lies more in the design of the i860 than in any one
compiler. If you really need it fast, write it in assembly for this chip.
There are a lot of subtle trade offs you make in coding for this chip that
are difficult to put into a compiler.


#######################################################################


>From ndoduc@framentec.fr Fri Sep 25 09:02:31 1992


AS far as f77 is concerned, I would suggest you to try:
1) MicroWay: ask for Yehia O'WEISS, phone=81 541 5466 in England (the US
      coordunates are in any Byte)
2) Lahey: Intel has asked them to write one compiler for the 860;
      email to: Dan Wright at lahey!support@uunet.UU.NET
there are others (portland group ... and possibly british ones too, but
I think MicroWay is the best)


--nh
Nhuan DODUC,
Framentec-Cognitech, Paris, France, doduc@framentec.fr
Association Francaise des Utilisateurs d'Unix, France, doduc@afuu.fr


#######################################################################


Reply-To: jhm@raster.mn.org (Jeff Michaud)


Hi Matthias,


A company called The Portland Group makes such a compiler for the 860. We
evaluated it, but decided the Metaware compiler was better suited to our
needs (mostly portability).


--
Jeff Michaud
jhm@raster.mn.org


#######################################################################


Reply-To: Andre Hinkenjann <hinkenja@jupiter.informatik.uni-dortmund.de>


Hallo Matthias,


ich schreibe gerade ein Betriebssystem fuer den i860. Dazu benutze ich den
Compiler der Portland Group. Das ganze wird von einer Firma gesponsert und
ich weiss, dass die sich vorher nach dem besten Compiler umgehoert haben
(bei Intel etc.). Dabei kristallisierte sich der Portland C-Compiler
heraus. Leider habe ich selbst noch keinen besonderen Wert auf die
Optimierungsfaehigkeiten gelegt, laut "Handbuch" sollen die aber ganz
immens sein. Irgendwo im Handbuch ist auch die mail-Adresse von Portland
angegeben (kann ich auf Wunsch auskramen). Allerdings meldet sich in der
newgroup comp.sys.intel (oder so) ergelmaessig jemand von Portland zu
Wort, den Du vielleicht anmailen koenntes.


Gruss,
      Andre
--

Return to the comp.compilers page.
Search the comp.compilers archives again.